What ingestion does
At install, Tweety Bird reads CLAUDE.md, AGENTS.md, and DECISIONS.md from the repo and writes their contents into that repo's shared memory. From the first text, Codex knows what your CLAUDE.md says and Claude Code knows what your AGENTS.md says.
- Your files are read, not edited.
- Run it again any time, for this or another repo:
tweetybird memory ingest
tweetybird memory ingest ~/code/apiSetup also adds one short, clearly marked block to your user-level instruction files (~/.codex/AGENTS.md for Codex) telling the agent that the shared memory exists and how to use it. tweetybird uninstall removes that block and leaves the rest of the file as it was.
What to keep in the file
- How to build, test, and run the project.
- Hard rules: never push to main, always use the design tokens, the commit author.
- Where things live in the repo.
These are true on every run and every agent reads them natively. Keep them short and current.
What to let the memory carry
- Decisions and the why behind them: "keep Stripe customer IDs denormalized on the org row because the join was killing the dashboard".
- Rejected approaches, so nobody proposes them again.
- What happened this week: the migration is half done, the flaky test is the webhook one.
This is the part that goes stale in a file, and the part memory keeps current automatically every turn.